sync(skills): mirror 9683ccec and pin @pathmode/mcp-server@1.38.0

.claude-plugin/plugin.json

@@ -1,7 +1,7 @@
 {
   "name": "pathmode",
   "displayName": "Pathmode",
-  "version": "0.1.44",
+  "version": "0.1.45",
   "description": "Run a deterministic preflight on your intent before an agent builds: /preflight scores six calibrated gates and names the exact blockers, no model call, no key. Pathmode also interrogates the intent behind a request (compile-intent), writes it to intent.md in your repo, and checks the code that comes back against the outcomes and constraints you agreed to. Free: uses the models you already have access to in Claude Code. Connect a Pathmode workspace to sync intent and evidence across a team.",
   "author": {
     "name": "Pathmode",

.mcp.json

@@ -2,7 +2,7 @@
     "mcpServers": {
         "pathmode": {
             "command": "npx",
-            "args": ["-y", "@pathmode/mcp-server@1.37.0"],
+            "args": ["-y", "@pathmode/mcp-server@1.38.0"],
             "env": {
                 "PATHMODE_API_KEY": "${user_config.api_key}"
             }

CALIBRATION.md

@@ -23,7 +23,7 @@ Three things get confused here, and they are not the same claim:
 3. **Whether the spec describes the right product decision.** **Nothing here measures this**, and
    no corpus of this shape could. A spec can score 6/6 and be a bad idea, clearly stated.
 
-Part A is a **regression baseline, not independent validation.** These 111 fixtures were written
+Part A is a **regression baseline, not independent validation.** These 192 fixtures were written
 alongside the rules they score, and the rules were tuned until they agreed. A high score is
 therefore evidence that the gate still behaves as it did when it was calibrated. It is not
 evidence that the gate generalizes. Independently collected, previously unseen specs would be
@@ -76,7 +76,7 @@ Each case records **where its defect actually shipped**, and they are not all eq
 | Case | Provenance |
 |---|---|
 | prose sections are content, not absence | shipped in `<= 1.26.3`, fixed in `1.27.0` |
-| non-English is unconfirmed, never absent | the known limitation the four-state verdict reports honestly |
+| an unread language is unconfirmed, never absent | the known limitation the four-state verdict reports honestly |
 | `## Why` is the objective, `## What` is not the outcomes | shipped in `<= 1.36.1`, fixed in `1.37.0` |
 | italic text the author wrote is content | never released; introduced and fixed between two commits |
 | a subsection heading is not a verification check | never released; same |
@@ -106,9 +106,10 @@ nothing was written.
 
 ## Known limitations, stated plainly
 
-- The gates match a **fixed English vocabulary**. A non-English objective cannot be confirmed, and
-  reports as `unconfirmed`. Part B has a Finnish case that holds this behaviour honest.
-- 111 fixtures is small, and they are **our own authored examples**, not sampled from real repos.
+- The gates match a **fixed English and Finnish vocabulary**. An objective in any other language
+  cannot be confirmed, and reports as `unconfirmed`. Part B has a German case that holds this
+  behaviour honest.
+- 192 fixtures is small, and they are **our own authored examples**, not sampled from real repos.
 - The gates judge **phrasing**, not correctness, novelty, feasibility, or whether the decision is
   a good one.
 - Part A cannot detect a document-parsing regression, as shown above.

README.md

@@ -72,9 +72,9 @@ For workspace API access without a local intent, you can still create an API key
 
 ## What's bundled
 
-**MCP server** — `@pathmode/mcp-server@1.37.0`, pinned so the plugin skills and server tool contract update together. Local mode with no key; cloud mode with one.
+**MCP server** — `@pathmode/mcp-server@1.38.0`, pinned so the plugin skills and server tool contract update together. Local mode with no key; cloud mode with one.
 
-**Check the gate yourself** — `node scripts/readiness-suite.mjs` runs the pinned server's preflight over 111 labelled field fixtures and a set of whole `intent.md` documents, and prints where it disagrees. Read [CALIBRATION.md](CALIBRATION.md) first: the field score is a regression baseline, not an accuracy claim.
+**Check the gate yourself** — `node scripts/readiness-suite.mjs` runs the pinned server's preflight over 192 labelled field fixtures and a set of whole `intent.md` documents, and prints where it disagrees. Read [CALIBRATION.md](CALIBRATION.md) first: the field score is a regression baseline, not an accuracy claim.
 
 **Command** — `/preflight` runs the deterministic six-gate readiness check and always ends in a verdict: on your `intent.md` if one exists, on a spec described in the conversation, or on a provisional draft it builds from context with assumptions marked.
 
@@ -109,9 +109,9 @@ completely silent in repos with no intent. If you would rather it did not run, r
 against. That tuning set is in this repo, mirrored from the monorepo that owns it, so you can check
 it rather than take the number on faith.
 
-[`readiness-corpus.json`](./readiness-corpus.json) holds 111 hand-labeled spec fragments, 58 labeled
-`good` and 53 labeled `vague`, spread across the gates: 24 objectives, 37 outcomes, 20 titles,
-18 constraints, 12 verification checks. Each item carries the text, the label, and the tags that
+[`readiness-corpus.json`](./readiness-corpus.json) holds 192 hand-labeled spec fragments (121 English, 71 Finnish),
+101 labeled `good` and 91 labeled `vague`, spread across the gates: 44 objectives, 78 outcomes,
+27 titles, 27 constraints, 16 verification checks. Each item carries the text, the label, and the tags that
 explain the call:
 
 ```json

fixtures/documents.json

@@ -68,10 +68,10 @@
     }
   },
   {
-    "name": "non-English is unconfirmed, never absent",
+    "name": "an unread language is unconfirmed, never absent",
     "file": "non-english-prose.md",
-    "why": "The gates match a fixed English vocabulary, so they cannot confirm this objective or constraint. That is a limit of the gate, and the verdict has to say so: 'unconfirmed' means it read the text and its word lists could not confirm it, which is different from claiming nothing was written. An audit of real specs found this class of failure is usually ours, not the author's.",
-    "provenance": "the known limitation the four-state verdict exists to report honestly",
+    "why": "The gates match a fixed English and Finnish vocabulary, so they cannot confirm this German objective or constraint. That is a limit of the gate, and the verdict has to say so: 'unconfirmed' means it read the text and its word lists could not confirm it, which is different from claiming nothing was written. An audit of real specs found this class of failure is usually ours, not the author's.",
+    "provenance": "the known limitation the four-state verdict exists to report honestly; the sample was Finnish until 1.38.0 started grading Finnish",
     "expect": { "objective": "unconfirmed", "constraints": "unconfirmed" }
   },
   {

fixtures/documents/non-english-prose.md

@@ -1,11 +1,11 @@
 ---
-title: "Parempi hakukokemus"
+title: "Bessere Sucherfahrung"
 ---
 
-# Parempi hakukokemus
+# Bessere Sucherfahrung
 
 ## Objective
-Käyttäjät eivät löydä tuotteita hakutoiminnolla, koska haku ei ymmärrä taivutusmuotoja.
+Nutzer finden keine Produkte über die Suche, weil die Suche keine Flexionsformen versteht.
 
 ## Constraints
-Ei saa hidastaa sivun latausta.
+Darf die Seite nicht verlangsamen.